StorageStorage that lives next to your GPUs
Two tiers, one workflow: keep hot working sets on the warm tier (Cluster Storage), park checkpoints and archives on the cold tier (Object Storage), and attach either to your workspaces and jobs.
Two tiers, matched to your data's temperature
The dataset you train on every day and last month's checkpoints have different needs. Pick the tier by how often the data gets touched.
Cluster Storage
High-availability distributed storage (CephFS on NVMe) bound to a cluster, the home for code, environments, and the datasets your active runs read every step.
- Shared over the cluster-local network, with no internet hop
- Read-write-many across concurrent workloads
- Data survives workspace stop, pause & terminate
- $0.20 per GiB-month, billed daily on actual usage
Object Storage
Durable S3-backed storage you can reach from any cluster in your organization. Park datasets, checkpoints, and long-term artifacts once. Attach them wherever the GPUs are.
- Accessible from any cluster in your org
- Checkpoints, logs & long-term artifacts
- S3-compatible access for external tools & pipelines
- $0.03 per GiB-month, billed daily on actual usage
From empty volume to mounted data
Three steps, from the web console or the CLI.
Create a volume
From the web console or vesslctl volume create: pick the warm or cold tier that fits the data.
Load your data
Upload from the console or vesslctl volume upload. Data lands once and stays ready for every run.
Attach and run
Mount volumes to workspaces or attach them to jobs at runtime. Multiple workloads can read and write the same volume.
Pay for what you store, daily
Billed daily on the data actually stored, not provisioned capacity.
Frequently Asked Questions
What's the difference between the warm and cold tiers?
The warm tier is Cluster Storage: high-availability CephFS-on-NVMe storage bound to one cluster, shared over the cluster-local network. It's best for code, environments, and the datasets your active runs read constantly. The cold tier is Object Storage: lower-cost, durable, S3-backed storage accessible from any cluster in your organization, best for checkpoints, logs, and long-term artifacts. Object Storage is slower, so don't use it as your main workspace path.
How is storage billed?
Both tiers are pay-as-you-go, billed daily on the amount of data actually stored, not on provisioned capacity. Cluster Storage is $0.20 per GiB-month (about $0.0067 per GiB-day) and Object Storage is $0.03 per GiB-month (about $0.001 per GiB-day), with rates effective July 1, 2026.
How do I create a volume and use it in a workload?
Object Storage volumes are self-service: create one from the web console or with vesslctl volume create, then load data from the console or with vesslctl volume upload. Cluster Storage volumes are set up by your organization admin and assigned to your team. Either way, mount volumes to workspaces or attach them to jobs at runtime. Both tiers support read-write-many, so multiple workloads can use the same volume at once.
Where should I keep training datasets and checkpoints?
Keep the working set your current runs read on the warm tier (Cluster Storage) for speed, and keep full datasets, checkpoints, and archives on the cold tier (Object Storage) so any cluster can reach them at lower cost.
Is there scratch space for temporary files?
Yes. Every workspace includes fast local scratch space (Temporary Storage) for caches and intermediate files. It's wiped when the workspace stops, so keep only data you can easily recreate there; anything you want to keep belongs on the warm or cold tier.
What happens to my data if my credits run out?
Your storage data is retained for up to 30 days with access restricted, then deleted. Top up before your balance reaches zero, and back up anything important beforehand.
Where AI models
get their GPUs.
- Start in minutes
- Scale to multi-node clusters
- Capacity reserved to your timeline
- Dedicated support